DAHLGREN, Va –
“This was an opportunity to help our scientists and engineers make sure they have the resources they need to solve hard problems promptly in ways we could not before.”
Ben Goldman, from Naval Surface Warfare Center Dahlgren Division (NSWCDD), is one of the co-leads for the intelligent automation thrust and recently assisted with the strategic planning and execution of the new DGX H100, known as the Naval Gold Standard Artificial Intelligence (AI) Computer.
The Naval Gold Standard AI Computer is a powerful complex system that accelerates the capabilities of NSWCDD scientists and engineers to process large quantities of data rapidly compared to a regular processing unit. The system, procured under the direction of the deputy technical director and the Technology Office, was supported by Naval Innovative Science and Engineering funding.
The workstation, located in the Innovation Lab (iLab), was built for AI machine learning (ML) processing and designed for complex graphics processing unit (GPU) computations. Computers typically have a central processing unit (CPU) and a GPU. CPUs are responsible for basic computational operations while GPUs excel at graphics and parallel computations. For processing AI/ML, a high-functioning GPU or even multiple GPUs are required to conduct graphics calculations which can quickly grow in complexity.
The internally facing capability will be used by the NSWCDD workforce primarily for technology exploration, development and maturation.
“We have assets and computational resources here at NSWCDD. However, the type of processing that our engineers and scientists were looking to do, even the basics, was taking days or weeks to achieve with how complex it was,” Goldman said.
“This system, and with the ones following it, will let us investigate, explore and build capabilities in less time. Things that would normally take multiple hours or even days, will take less than an hour. If we are trying to explore what the best solution to a problem is and try 30 different options rather than three, we have a higher chance of succeeding and getting better performing systems for the users.”
End user and AI/ML researcher Michelle Bolner is one of many NSWCDD scientists and engineers looking forward to the speed of the new system and the rapid procurement of data it will produce.
“Setting up your environment and ensuring you have technology available that can handle the computations can take months out of your time,” Bolner said. “You can’t afford to spend months setting up your environment and ordering assets that are powerful enough for you to use. Having a DGX, or even multiple, is a huge deal for us.”
The NSWCDD team utilized a Cooperative Research and Development Agreement with NVIDIA as well as worked with other supporting partners to successfully locate the hardware that was right for the problem set, which expedited the overall process.
“These assets are in huge demand. All the big industry companies are trying to buy computers like this,” Goldman said. “We were buying one, but they are buying thousands, so because of our partnerships with companies like NVIDIA, we were able to make this happen.”
NSWCDD operation design and performance engineer Dr. Tyler Ferro, assisted with the coordination between NVIDIA, NSWCDD personnel and other involved parties. He was also responsible for the iLab coordination and facilitating communication.
“While time may be the key factor in the benefits the system provides, one element that cannot be overlooked is the cost-saving component,” Ferro said.
“The end users can use a minimalistic computer to get the basics of their work done and then send the rest of it to the server to finish off. So we are saving a considerable amount of time and money because not every individual needs a high-powered computer of this capacity.”
iLab IT manager Eric Hayes and former iLab engineer Michael Darnell also worked on the procurement team for the acquisition of the DGX H100 system.
“What made this acquisition so different was we didn’t already have technology like it at NSWCDD,” Hayes said. “This was new to everyone so we definitely shared responsibilities and it was a lot of teamwork.”
“The original idea was to have something readily available to the NSWCDD workforce, so that was why it fell under the iLab,” Darnell said. “We wanted a place for it that it was easily accessible to all the departments.”
The system is still undergoing final testing to ensure it is functioning at maximum capacity. “This was a strategic investment for the command, so it is something we want to make sure we have set up and configured to run at its peak speed and also be easily accessible to the scientists and engineers that are going to be interacting with it,” Goldman said.
One major benefit the system will provide is the multi-departmental collaboration. End-user Molly Thomson, who is with NSWCDD, will be utilizing the system for model training and is excited about the advantages of collaborating with other departments.
“This could be beneficial because there could be several groups who need similar data and similar models. So having that shared resource is vital so we can all work together to solve those complex problems,” Thomson expressed.
Imaging scientist from NSWCDD Eric Montag is looking forward to running multiple tests at the same time with a faster end result.
“We can train multiple models at the same time with this new capability and get the results, and then make a decision on which one to go with,” Montag said. “However, because there is more computing power overall, we can run multiple different model trainings at the same time. We can do it all on one computer and faster, instead of using multiple different computers.”
“I feel like people need to realize that we are using cutting-edge technology,” Thomson said. “This kind of workstation is one of the best you can get, and this is what big industry uses. People should know that we are very much in the game, and that’s exciting.”